The revised manuscript is thorough. My comments and questions in the first review are addressed in the revised manuscript and/or the interactive discussion. The conclusion about the time reversal asymmetry became modest in the revised manuscript because of a new result (Fig. 4). Nevertheless, the methodology and its application results in this article will be useful for future researches of inverse stochastic-dynamic modelling of abrupt climate changes. I recommend the publication of this article.
Possible technical corrections:
The following point in Page 3, Line 8 might be misleading: "Bayesian framework to compare different types of models has 25 also recently been used for the specific case of the NGRIP δ18O record, including a double-well potential model, a relaxation oscillator, and two versions of a mixture of locally linear stochastic models. Based on the Bayesian information criterion, it was concluded there that the relaxation oscillator model is best supported by the observations (Kwasniok, 2013)."
In Kwasniok 2013, two versions of a mixture of locally linear stochastic models are more favored than the relaxation oscillator on the basis of BIC (thought he mentions that those stochastic linear models correspond to the phases of the relaxation oscillation).
Page 6, Line 27: \mathbf{x_n} is \mathbf{x}_n?
Both r and r^P are used for the Person's correlation coefficient.
In the caption of Figure 5, "(A) Difference" should be "(E) Difference". |